Learnable Speed Control for Speech Synthesis

1. A method of synthesizing speech at one or more speeds, comprising: encoding, by a computer, a context associated with one or more phonemes corresponding to a speaking voice; aligning, by the computer, the one or more phonemes to one or more target acoustic frames based on the encoded context; recursively generating, by the computer, one or more mel-spectrogram features from the aligned phonemes and the target acoustic frames; and synthesizing, by the computer, a voice sample at a given speed corresponding to the speaking voice using the generated mel-spectrogram features wherein the encoding comprises: receiving a sequence of the one or more phonemes; and outputting a sequence of one or more hidden states containing a sequential representation associated with the received sequence of phonemes, and wherein the aligning the one or more phonemes to one or more target acoustic frames comprises: concatenating the output sequence of hidden states with information corresponding to the speaking voice; applying dimension reduction to the concatenated output sequence using a fully connected layer; expanding the dimension-reduced output sequence based on a rate associated with each phoneme; and aligning the expanded output sequence to the one or more target acoustic frames.

2. The method of claim 1 , further comprising concatenating one or more frame-aligned hidden states with a frame level, a root mean square error value, and a relative position associated with every frame.

3. The method of claim 2 , wherein the rate of each phoneme is obtained from a force alignment performed on one or more input phonemes and one or more acoustic features.

4. The method of claim 1 , wherein the generating the one or more mel-spectrogram features based on the aligned frames comprises: computing an attention context from one or more encoded hidden states aligned with the one or more target acoustic frames; and applying a CBHG technique to the computed attention context.

5. The method of claim 4 , wherein a loss value associated with the mel-spectrogram is minimized.

6. The method of claim 1 , wherein the generating the one or more mel-spectrogram features is performed by a recursive neural network.

7. The method of claim 6 , wherein the inputs to the recursive neural network comprise the sequence of the one or more phonemes, a rate associated with each of the one or more phonemes, a root mean square error value, and an identity associated with a speaker.

8. The method of claim 1 , wherein the voice sample is synthesized without parallel data and without changing content associated with the speaking voice.

9. A computer system for synthesizing speech at one or more speeds, the computer system comprising: one or more computer-readable non-transitory storage media configured to store computer program code; and one or more computer processors configured to access said computer program code and operate as instructed by said computer program code, said computer program code including: encoding code configured to cause the one or more computer processors to encode a context associated with one or more phonemes corresponding to a speaking voice; aligning code configured to cause the one or more computer processors to align the one or more phonemes to one or more target acoustic frames based on the encoded context; generating code configured to cause the one or more computer processors to recursively generate one or more mel-spectrogram features from the aligned phonemes and the one or more target acoustic frames; and synthesizing code configured to cause the one or more computer processors to synthesize a voice sample at a given speed corresponding to the speaking voice using the generated mel-spectrogram features, wherein the encoding code comprises: code configured to cause the one or more computer processors to receive a sequence of the one or more phonemes; and outputting code configured to cause the one or more computer processors to output a sequence of one or more hidden states containing a sequential representation associated with the received sequence of phonemes, and wherein the aligning code comprises: concatenating code configured to cause the one or more computer processors to concatenate the output sequence of hidden states with information corresponding to the speaking voice; applying code configured to cause the one or more computer processors to apply dimension reduction to the concatenated output sequence using a fully connected layer; expanding code configured to cause the one or more computer processors to expand the dimension-reduced output sequence based on a rate associated with each phoneme; and second aligning code configured to cause the one or more computer processors to align the expanded output sequence to the one or more target acoustic frames.

10. The computer system of claim 9 , wherein the concatenating code configured to cause the one or more computer processors to concatenate one or more frame-aligned hidden states with a frame level, a root mean square error value, and a relative position associated with every frame.

11. The computer system of claim 10 , wherein the rate of each phoneme is obtained from a force alignment performed on one or more input phonemes and one or more acoustic features.

12. The computer system of claim 9 , wherein the generating code comprises: computing code configured to cause the one or more computer processors to compute an attention context from one or more encoded hidden states aligned with the one or more target acoustic frames; and applying code configured to cause the one or more computer processors to apply a CBHG technique to the computed attention context.

13. The computer system of claim 9 , wherein the generating the one or more mel-spectrogram features is performed by a recursive neural network.

14. The computer system of claim 13 , wherein the inputs to the recursive neural network comprise the sequence of the one or more phonemes, a rate associated with each of the one or more phonemes, a root mean square error value, and an identity associated with a speaker.

15. The computer system of claim 9 , wherein the voice sample is synthesized without parallel data and without changing content associated with the speaking voice.

16. A non-transitory computer readable medium having stored thereon a computer program for synthesizing speech at one or more speeds, the computer program configured to cause one or more computer processors to: encode a context associated with one or more phonemes corresponding to a speaking voice; align the one or more phonemes to one or more target acoustic frames based on the encoded context; recursively generate one or more mel-spectrogram features from the aligned phonemes and the one or more target acoustic frames; and synthesize a voice sample at a given speed corresponding to the speaking voice using the generated mel-spectrogram features, wherein the computer program is further configured to cause the one or more computer processors to: receive a sequence of the one or more phonemes; output a sequence of one or more hidden states containing a sequential representation associated with the received sequence of phonemes; concatenate the output sequence of hidden states with information corresponding to the speaking voice; apply dimension reduction to the concatenated output sequence using a fully connected layer; expand the dimension-reduced output sequence based on a rate associated with each phoneme; and align the expanded output sequence to the one or more target acoustic frames.